In maritime operations, instances occur where vessels flee after colliding with buoys. Automatic Identification System (AIS) data collected from these buoys play a vital role in identifying the responsible party and determining liability. However, data from buoys can be lost in rough sea conditions and equipment failure, which makes it difficult to identify accidents where ships collide with buoys. To address this issue, this study proposes a model that combines Long Short-Term Memory Network (LSTM) with Multiple Attention mechanisms to reconstruct buoy trajectories with high precision. In this paper, the buoy trajectory reconstruction test is carried out using Qingdao Harbour Buoy 304 a case study, and the validation results show that the mean squared error (MSE) of this model is 12.15. In addition, the LSTM-Attention model shows a significant improvement in all the metrics compared with other models: compared with the Random sampling model, the mean absolute error (MAE) is improved by 75.15%, and the root mean squared error (RMSE) by 71.02%; MSE by 77.31%, MAE by 36.70%, and RMSE by 52.36% compared to the Cubic spline model; and MSE by 49.92%, MAE by 27.35%, and RMSE by 29.52% compared to the Hermite model. These results show that the LSTM-Attention model significantly improves the accuracy and reliability of trajectory reconstruction.
Read full abstract